Introduction¶
In the DeepSeek Harness (DSH) plugin ecosystem, adding image processing capabilities to models typically requires intercepting specific message streams. dsh-pro-vision-std is a rewritten component based on the dsh-std specification, providing a lighter and more standard approach. This plugin decouples image processing logic from the reasoning logic of the V4-Pro model. It supports image attachments by first calling a vision model to generate descriptions and then using V4-Pro for text generation.
Core Features¶
- Portable ModelProvider: As an independent ModelProvider, it enables DeepSeek V4-Pro to handle image attachments without modifying the core logic of DSH.
- Image Processing Workflow: Uses the
deepseek-v4-flash-vision-expmodel to generate precise textual descriptions of images, then sends the generated descriptions todeepseek-v4-profor the final response. - Contract Integration: Uses the
dsh-std readImagecontract to read images and publishes thedsh-std Tool describe_imagetool for explicit invocation. - In-Process Caching: For identical image byte streams, the plugin performs in-process caching to reduce overhead from repeated processing.
Installation and Activation¶
Installing this plugin requires first installing its dependent adapter @dsh-std/adapter-dsh.
dsh plugin --profile web add @dsh-std/adapter-dsh
dsh plugin --profile web add github:lasdrder0705/dsh-pro-vision-std
After installation or update, the corresponding DSH Profile must be restarted for changes to take effect. The update command is as follows:
dsh plugin --profile web update dsh-pro-vision-std
Typical Usage¶
In the DSH model selector, select the following route to enable text + image capability:
dsh-pro-vision / deepseek-v4-pro
dsh-pro-vision / deepseek-v4-pro-vision is a compatibility alias for the above route and has the same functionality.
If only pure text capability is needed, select the official route:
deepseek-official / deepseek-v4-pro
The plugin automatically reuses the already resolved DEEPSEEK_API_KEY from the DSH Credential Service, so no re-entry is required.
For configuration, optional environment variables can be used for adjustments:
| Variable Name | Default Value |
|---|---|
DSH_PRO_VISION_BASE_URL |
https://api.deepseek.com |
DSH_PRO_VISION_VISION_MODEL |
deepseek-v4-flash-vision-exp |
DSH_PRO_VISION_TEXT_MODEL |
deepseek-v4-pro |
DSH_PRO_VISION_MAX_VISION_TOKENS |
8192 |
DSH_PRO_VISION_REASONING_EFFORT |
high |
DSH_PRO_VISION_CACHE_SIZE |
200 |
Applicable Scenarios and Notes¶
- Compatibility Requirement: This plugin is compatible with
dsh-std Community v0.15. - Environment Dependencies: The
@dsh-std/adapter-dshadapter must be installed, and the runtime environment must support Node.js 22.19+ or 24+. - Isolation: Installing this plugin does not affect or impersonate the official
deepseek-officialprovider. DSH isolates different Providers through routing, ensuring that plugins do not interfere with one another. - Security and Logging: Credential values are resolved only at runtime and are not written to the plugin configuration file or recorded in logs.
Brief Conclusion¶
dsh-pro-vision-std provides a standardized method for DeepSeek V4-Pro to process image input through image descriptions. It is independent of the original plugin and conforms to the dsh-std specification, making it suitable for DSH users who need to integrate image capabilities while maintaining clean code and compatibility. For more details, refer to the Community Catalog or the Project Repository.